[Xorg] The big multiconsole nasty

[Egbert Eich]

Alan Cox writes:
 > Well actually there are two or three related issues here
 > 
 > 1. Multiple video cards means making RAC work across *multiple* X
 > servers running at once on a system each with its own console

Right. That is currently the biggest showstopper to get multi seat 
working. One could use an external deamon to handle locking between
servers or a kernel module.

 > 
 > 2. PCI config has to go via the kernel, including some kind of
 > arbitration for VGA enables. Bashing PCI config ports directly will
 > crash machines and (I'm told) also lead to memory and I/O corruption 
 > on AMD64 IOMMU..

Do you refer to config port access directly thru the 'well known' 'config
mechanism 1' (ie PIO 0xcf8 0xcfc registers), or do you mean access to
PCI config space from user space altogether - even when using the kernel
functions?

 > 
 > 3. Borrowing things like BIOS mode switch functionality is going to
 > get really hairy.
 > 
 > Has anyone in the original design of the resource arbitration logic
 > considered the multi-server issue ?

Yes - after we have gotten thru with the present design :-/
I have spent some time and thoughts on this issue:
My conclusion was that one should move the entire device and 
chipset probing (currently Probe() and PreInit() in the X drivers)
to an external deamon. This would also take care of the access control
among different Xserver instances. 

Cheers,
	Egbert.